Target imgs: attached, broken, threshold, blob (redefined)

<CC attaching>
39_192
43_196
28_181
17_70
14_73
6_91
3_140
1_114
310_130-138, 99-103
309_145
309_136
307_153
306_153
304_81
303_81
301_159
284_184
283_185
276_192
293_174-173
258_205
257_206
238_216
229_220
210_229
29_182

<CC broken>
305_81
304_81

<Thresh problem>
302_74
295_174
288_54
287_55
286_52
279_47
99_227
97_225
94_12
48_34

<blobs> no angular neighbor in current FOV
229_220
29_182
238_216
257_206
258_205

CC detach, filtering

1. dealing with remaining CC after noise removal
1. do we re-index the obj so that the indices are an array of continuous numbers?
2. this means to re-index each pixel in the matrix, which requires time
3. we can use a dict to record the remaining indices in the CC map, which takes space of only 10- elems
4. go with dict !!!!
2. Matlab boundary detection
1. [B,L,N,A] = bwboundaries(BW);
2. % B = {obj, holes}
3. N = n_obj
4. A = [boundaries, r vs. c =  enclosing(i_obj_in_B) vs. enclosed(i_hole_in_B)]

top as CC or bottom as CC?

Prerequisite

• For any edges of a CC
• the inner edge (of shorter radius) is indexed as (1), outer (2)

Top:

• Pros
• more direct and intuitive for CC analysis
• no Inf in the data range, and they are really "connected".
• edges are valid depth data
• when calculating depth of the groove bottom, the depth of the top is directly available.
• Cons
• the edges are not the boundaries of the same groove, making the forthcoming groove locating more complex.
• to locate groove(k), we have to get CC(k+1).edge(1) and CC(k).edge(2)
• with another orientation (e.g. 12 o'clock), it might change into CC(k+1).edge(2) and CC(k).edge(1)

Bottom

• Pros
• more direct and intuitive for later polar coord extraction for grooves.
• always CC(k).edge(1), CC(k).edge(2)
• Cons
• the CC analysis is less intuitive, need to combine the sidewall and the bottom as a pseudo-CC.
• no valid depth value on the edge
• when addressing the depth of the bottom, the reference depth of the top should be derived from surrounding area. This require a branched navigation to find adjacent valid area,
• At 10 o'clock, go from 5-10 o'clock for edge(2)
• At 2 o'clock, go from 8-2 o'clock for edge(2)

test image

rewrite Matlab CC analysis

Description

• Matlab CC analysis assumes that the image can be loaded into memory. But in our case, stitched matrix exceeds the memory limit. We cannot do a CC analysis of the grid using Matlab native function.

What I want

• Trace the groove smoothly across FOVs without hassling at the boundary.

How I can have them

• Mark the areas in the grid as CCs, link all CCs in all frames
• When crossing the boundary of two adjacent frames, find the CC Y in FOV B that is the continuation of the CC X in FOV A
• This can be a round trip, because all CCs are essentially one. So the output of this step is a chain like: Obj 1 in FOV-1, Obj 4 in FOV-2, ....., Obj 3 in FOV-2, Obj 6 in FOV-1.
• Exceptions: a node in this chain can have several FOVs (overlaps). So the chain is actually a tree.
• How to build the tree?
• Global: need a head and top-down
• Pros: conceptually easy.
• Cons: directional, need decision making about the spread direction.
• Local: for each FOV, find its precedents and successors
• Pros:
• Cons: still need to determine which FOVs are precedents / successors, even harder than global approach, because there is no "spreading" at all, so the direction is N/A.
• What will avoid direction decision?
• Angle, with the center coord, we can determine the angles of the entry/exit points.
• Select one of them as a current obj of interest and trace through it, collecting radial distances and depths.
• When I collect radial distances of the target positions, I don't have to deal with boundary again, just go with a label map
• find all pixels with a label "i".
• calculate the radial distances without having to care about the order of these pixels.
• we can sort the pixels in terms of the angular order later.

Any assumptions associated with this workflow and any exceptions

What I need to deal with the exceptions and achieve the original goal

What I have

What I don't have

What I shall do next